Assessing serum thrombopoietin for enhanced diagnosis of ITP, AA, and MDS using machine learning: A retrospective cohort study.

Differentiating between immune thrombocytopenia (ITP), aplastic anemia (AA), and myelodysplastic syndromes (MDS) is critical due to the distinct treatment approaches required for each condition. This study investigates the role of serum thrombopoietin (TPO) levels as a potential biomarker to aid in the diagnosis of these hematological disorders. This retrospective cohort study analyzed serum TPO levels in patients diagnosed with ITP, AA, and MDS, using clinical records and stored serum samples collected from patients treated between September 2023 and May 2024. Statistical analyses were performed to determine cut-off values for TPO levels that effectively differentiate between these conditions. Additionally, machine learning models were utilized to enhance diagnostic accuracy based on clinical indicators, including TPO levels. Serum TPO levels were markedly elevated in AA (1369.19 ± 751.26 pg/ml) compared to ITP (263.57 ± 355.91 pg/ml), MDS (434.55 ± 551.56 pg/ml), and health control (71.64 ± 30.32 pg/ml) (P < 0.0001). Correlation analysis revealed a significant positive correlation between TPO levels and ITP, AA, and MDS (P < 0.0001), Linear regression analysis indicated that age was a significant predictor of TPO levels (P < 0.0001). The optimal cut-off value for TPO levels distinguishing ITP from AA was 302.43 pg/mL, yielding an AUC of 0.925 (sensitivity with 80.75%, specificity with 94.06%). Machine learning models demonstrated that Logistic Regression, XGBoost, and LightGBM performed best, with the Logistic Regression achieving an accuracy of 86.3% and an AUC of 0.910. Serum TPO levels are a promising non-invasive biomarker for distinguishing between ITP, AA, and MDS. Incorporating TPO measurements into clinical practice may enhance diagnostic accuracy and improve patient management strategies.